Page 1
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-43-
A Proposed Data Warehouse Framework to Enhance Decisions
of Distribution System in Pharmaceutical Sector
Noura Mahmoud Abd Elazeem, Nevine Makram Labib, Aliaa Kamal Abdella
Computer and Information Systems Department Sadat Academy for Management Sciences, Cairo, Egypt
[email protected] , [email protected] , [email protected]
Abstract
Noncommunicable diseases, including cardiovascular diseases, diabetes, cancer, hypertension,
and chronic respiratory diseases, are currently the leading national cause of death in Egypt. There are many problems facing the pharmaceutical companies such as increasing the cost of inventory and
expired medicines. Recent trends of Business Intelligent have taken attention of both communities and
research in pharmaceutical sector. Data warehousing techniques used to enhance decisions of
pharmaceutical distribution department by predicting the sales of medicines. Auto regressive moving average model time series and Neural Networks are used to predict sales based on historical data.
Keywords: Business Intelligence, Data Warehouse, Pharmaceutical Distribution, Decision
Support, Auto Regressive Moving Average, Neural Networks.
1. Introduction
Noncommunicable diseases (NCDs) are estimated to account for 82% of all deaths in
Egypt and 67% of premature deaths. The 2011/2012 STEPwise survey, conducted by the
Ministry of Health and Population, in collaboration with World Health Organization(WHO),
revealed a significantly high prevalence of risk factors for NCDs among the adult population,
including: a 24% prevalence of smoking and a growing use of shisha tobacco, one of the most
overweight populations in the world, with 66% of women overweight , 42% obese and almost
three quarters of the population not involved in vigorous activity, 17% prevalence of diabetes
and 40% prevalence of hypertension. Egyptians have an average daily salt intake of 9 grams,
nearly double the recommended allowance [1]
Data storage and information retrieval is a very important topic nowadays and affects a
large number of people and economic agents, being a valuable source for decision making or
increasing business. The efficient and effective use of information is particularly important in
Business Intelligence (BI). It is important to understand the relationships between different
aspects of the company to be derived towards specific objectives such as increasing the market
share and improving customer satisfaction. Business Intelligence is critical in supporting
decisions. This type of solution is due to the fact that companies are drowning in data that record
in operational databases: payroll data, financial data, customer data, vendor data, and so on.
These databases are typically tuned for each operation, such as retrieving a single customer
order, or for specific batch jobs, such as processing payroll at the end of each month. These
databases are not designed to communicate with one another, allowing users to explore data in
an unusual way, or to provide high level summary data at once [2].
Data warehousing (DWH) is popular in business environments and encapsulates the
process of transforming and aggregating operational data and bringing it to a platform
optimized for efficient storage and advanced analysis [3]. Data warehouse has offered an
excellent solution towards right decisions in pharmaceutical companies. A Data Warehouse
Page 2
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-44-
offers a solution designed to enable these enterprises to easily obtain relevant, accurate and up-
to date information about prescribers, managed care organizations, wholesalers, distributors
and consumers. Pharmaceutical companies have a growing need to combine the large amounts
of their downstream data such as supply chain or inventory information with their abundance
of sale data. This data must be combined for a clear picture of the supply chain to be integrated
into a data warehouse. A Data Warehouse allows companies to confidently make product and
research decisions based on integrated, detailed product and portfolio life cycle data. Integrated
data from across the company allows pharmaceutical firms to determine the types of drugs to
focus its research initiatives on. The pharmaceutical firm can make “right time” decisions by
analyzing data from the DWH. It can also gain insight into its market share as well as those of
its competitors [4].
Enhancing sales and operations planning through forecasting analysis and business
intelligence is demanded in any industry and business. Sales forecasting, in pharmaceutical
distribution companies plays a major role for enterprises in making business plans more
accurate and gaining competitive advantage. Data mining (DM) methods are used to analyze
large observational data sets, find unsuspected relationships, and discover patterns and trends [5].
The researcher collects from Egyptian Company for Medicine Trade, historical data for
ten years about diabetic and hypertension drugs to build databases (DBs) in DWH. Data mining
techniques such as Neural Networks (NN) and time series were applied for sales prediction to
enhance decisions in pharmaceutical sector.
2. Objective of The Study
The main objective of the study is to propose a data warehouse framework to enhance
decisions of distribution systems in pharmaceutical companies to decrease the medicine
industry cost and increase the productivity.
3. Previous Studies
Egypt was ranked 8th highest country in the world in terms of diabetes rates in 2013. The
prevalence of diabetic in Egypt was found in around 15.6% of all adults aged 20 to 79 in 2015.
The World Bank reported an even higher percentage (16.7%). Today, more than 7.8 million
Egyptians suffer from diabetic and this number is expected to double by 2035 [6].
A number of people with diabetes from 2005 to 2014 shown in Figure1:
Figure1. Diabetes Rates in Egypt
Hypertension is a chief public health care in both developing and developed countries.
Hypertension affects approximately 1 billion individuals worldwide. Egypt was ranked 4th
highest country in the world and the first one in Africa in terms of Hypertension rates.
Page 3
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-45-
Hypertension is an urgent health problem in Egypt with prevalence rate of 26.3% among the
adult population. Its incidence increases with aging, around 50% of Egyptians over the age of
60 years have hypertension [7].
A number of people with hypertension from 2005 to 2015 shown in Figure2:
Figure2. Hypertension Rates in Egypt
The amount of expired medicines in 2017 was about 600 million EGP for diabetic,
hypertension and cardiovascular medicines in Egypt, according to the Ministry of Health [8].
The techniques of data mining and data warehouse used for prediction in pharmaceutical
and healthcare industry that shown in Table 1:
Table1.Techniques Used in Pharmaceutical and Health Care Sector
Domain Name of Paper Techniques
Predict sales of pharmaceutical distribution
companies
Intelligent Sales Prediction for
Pharmaceutical Distribution
Companies: A Data Mining Based
Approach [9]
Network analysis tools and time series
forecasting methods
Drug consumption forecasting in pharmaceutical
industry production planning
Application of Data Mining
Techniques in Drug Consumption
Forecasting to Help Pharmaceutical Industry Production Planning [10]
Artificial Neural Networks (ANN) and
Decision Tree (DT)
Healthcare domain to diagnosis diabetes disease
An Expert Clinical Decision Support System to Predict Disease Using Classification Techniques [11]
Decision Tree and K-
Nearest Neighbor
(KNN)
Healthcare domain to
analyze diagnosis and
treatment of Breast Cancer
disease
Data Mining Techniques in Health Informatics: A Case Study from
Breast Cancer Research [12]
Decision Tree
Healthcare domain to predict kidney diseases
Data Mining Techniques for the
Prediction of Kidney Diseases and
Treatment [13]
Decision Tree
Healthcare domain to predict
diabetes
Performance Analysis of Data Mining Classification Techniques
to Predict Diabetes [14]
Decision Tree
Healthcare domain in cancer
diseases.
The Technology of Using A Data Warehouse to Support Decision-
Making In Health Care [15]
DWH, OLAP
Healthcare domain in
influenza diseases.
Health Care Data Warehouse
System Architecture for Influenza (Flu)Diseases [16]
DWH, OLAP
Page 4
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-46-
There are many techniques used in healthcare and pharmaceutical sector that show in
previous studies table, such as: OLAP, Decision tree, ANN, Time series, K-NN, Classification
and Clustering. Most of the studies used decision tree, ANN and time series because they
proved that these techniques give the best results in prediction. So, the researcher found that
ANN and time series are efficient techniques for the historical data of Egyptian Company for
Medicine Trade because it has no correlation between its variables and it has non-linear trend.
4. Data Warehouses in Healthcare
A framework is a conceptual or actual structure prepared to serve as a conductor or
support for building of something that extends the structure into something useful [17]. The
main goal of the proposed framework was to enhance decisions in the distributed
pharmaceutical company based on sales prediction. The prediction techniques were chosen
based on performance evaluation.
4.1. Framework Description
The framework can be described in four phases that shown in Figure3. Phase one is
consisted of data preparation phase which has four steps (data collection, building DBs, DWH
and data cleaning). Phase two is consisted of training phase which is applying time series to
three types of Neural Networks techniques (levenberg marquardt, Bayesian regularized, and
Scaled conjugate gradient).Phase three is testing the performance based on mean square error
(MSE). Phase four is consisted about evaluating the performance of the best prediction model.
Figure3. The Proposed Framework of Sales Prediction
4.2. Data Warehousing Phase
There are four steps to design and implement DWH that can be shown as the following:
Page 5
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-47-
4.2.1. Data Collection
Data were collected from Egyptian Company for Medicines Trade by gathering invoices
about Diabetic and Hypertension drugs for the last ten years from 2008 to 2017. There are
twenty different brands of Diabetic drugs with different amount of sales for each year and thirty
different brands of Hypertension drugs with different amount of sales for each year.
4.2.2. Building Databases
Database is one of the main components of the Information system (IS). The main goal
of IS is to transform data into information which used to produce knowledge needed for
decision making. The information system DB’s purpose should be able to take data and provide
tools for aggregation and analysis to help for decision making. Database is
an organized collection, because in a database, all data is described and associated with other
data [18]. Databases organize in many different ways and take many forms. The most popular
form of DB today is the relational DB. A relational DB is a set of described tables from which
data can be accessed or grouped in many different ways without having to reorganize the DB
tables [18].
The relational DB for pharmaceutical company can be showed in five tables with their
relationships in Figure4 as the following:
Figure 4. Relational Database
Invoice_dt table has attributes:
Invoice_no (with a primary key)
Invoice_type
Item_cd
Qty
Unit_price
This table has “one to many” relationship with table called Item that has attributes:
Item_cd (with a primary key)
Item_nm_en
Item_nm_ar
Prs_price
Sal_price
Category_id
Page 6
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-48-
And Invoice_dt table has indirect relationship with Invoice_type table that has attributes:
Invoice_type(with a primary key)
Invoice_type_nm
And Invoice_hd table has attributes:
Invoice_no
Invoice type
Invoice_date
Tot_invoice
Finally, Category table has “one to many “relationship with Item table and has two
attributes:
Category_id (with a primary key)
Category_name
4.2.3. Data Warehouse
The purpose of DWH is to take large data from heterogeneous sources and prepare
them in known formats that helps in understanding and for making decisions. Data warehouse
is providing direct access by using graphical tools for querying and reporting [19, 20]. The data
about diabetic and hypertension drugs was collected together which relating to various DBS of
individual years at last ten years as shown in Figure5:
Figure5. Schematic Diagram for Data Warehouse
4.2.4. Data Cleaning
Data cleanup is the process of looking for and fixing inconsistencies to ensure that data
is accurate and complete [21]. The main purpose of data cleaning is to detect, correct errors and
inconsistencies from data to develop data with its characteristics. The characteristic of data is
to be complete, accurate, accessible, economical, flexible, reliable, relevant, simple, timely,
verifiable, and secure. It is confirmed to be a difficult but unavoidable task for any IS [22].
DB
2008 DB
2009
DWH
DB
2017
DB
2016 DB
2015
DB
2014 DB
2013 DB
2012
DB
2011
DB
2010
DB
2008
DB
2009
DB
2017
DB
2016 DB
2015
DB
2014 DB
2013
DB
2012
DB
2011
DB
2010
Diabetic Drugs Databases
Hypertension Drugs Databases
Page 7
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-49-
The Items, Invoice_dt and Category tables are determined to select most significant
attributes (quantity, category_name, and item_name_en) for prediction in data mining
techniques.
4.3. Applying Time Series Model
A time series is one of predictive data mining techniques. It is a set of numbers that
measures the status of some activity over time. It is also a collection of data recorded over a
period of time, weekly, monthly, quarterly, or yearly. Time series equation
(zt=y*t−y*t−1−y*t−L +y*t−L−1) [23].
4.3.1. Time Series Forecasting in Diabetic Drugs
Mean squared error is calculated as the average of the forecast error values [24]. The
results of applying time series in diabetic drugs data at the last ten years show in Table 2 and
Figure6:
Table2. Time Series Forecasting for Diabetic Drugs
Years Mean Squared Error
2008 470074.75
2009 647949.75
2010 645752.25
2011 617818.0625
2012 626656
2013 1139090.04
2014 1773367.4375
2015 2159805.25
2016 2815866
2017 2886640.54
The results showed that the best predicated year was in 2008 that have the lowest mean
squared error.
Figure 6. Classifier Model in Diabetic Drugs Data
There is no correlation coefficient between variables as shown in this figure.
4.3.2. Time Series Forecasting in Hypertension Drugs
Page 8
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-50-
The results of applying time series in hypertension drugs data at the last ten years show
in Table 3 and Figure7:
Table3.Time Series Forecasting for Hypertension Drugs
Years Mean Squared Error
2008 226693.5378
2009 218138.2233
2010 211518.037
2011 184188.3363
2012 194010.8619
2013 204505.3611
2014 211581.9167
2015 216821.1389
2016 215795.5285
2017 221755.4563
The results showed that the best predicated year was in 2011 that have the lowest mean
squared error.
Figure7. Classifier Model in Hypertension Drugs Data
There is no correlation coefficient between variables as shown in this figure.
There are observations from the previous results that can be conducted as follows:
1. The future prediction of trained data is nonlinear for both diabetic and hypertension drugs
as shown in Figure8 and Figure9:
Figure8. Train Future Prediction in Diabetic Drugs Data
Page 9
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-51-
Fig.9. Train Future Prediction in Hypertension Drugs Data
2. Time series are very complex because each observation is dependent upon the previous
observation [25].
3. Random error is influential from one observation to another. These influences are called
autocorrelation dependent relationships between successive observations of the same
variable [25].
4. Focus on univariate data with linear [25].
4.4. Applying Neural Networks
Neural Network is known as a computing system that contained of a number of simple,
highly interconnected processing elements, which it can processing information by their
dynamic state to external inputs. The Neural Networks using a dynamic network and helps users
to selecting data, training, validation, and testing sets, and training the network [26].
4.4.1. Neural Networks Models
There are three models to a high order Neural Networks at MATLAB:
1. The Levenberg-Marquardt (LM) is a numerical least-squares non-linear function
minimization technique. The LM computes the weight change according to:∆w = (JT(w)J
(w) +µI) -1JT(w)e(w) [27].
2. Bayesian regularized Neural Networks (BRNNs) restricts the magnitude of the weights by
using the equation: P(A|B) = P(B|A) P(A)/P(B) [28].
3. Scaled Conjugate Gradient adds to the complexity of the training procedure by performing
a line search in each iteration to find the best step size along the conjugate direction. Instead
of using a line search, the scaled conjugate gradient method uses a Levenberg-Marquardt
approach to determine the optimal step size at each iteration by using the equation: pt+1 =
rt+1 +βtpt [29].
4.4.2. Applying Neural Networks in Diabetic Drugs
1. Applying LM Technique in Diabetic Drugs
Levenberg-Marquardt is a Neural Networks training function that updates weight and bias
values of the Neural Network. LM technique is often the fastest backpropagation technique,
and is highly recommended as a first choice supervised technique for training moderate sized
(up to several hundred weights) feed-forward Neural Network [27].
Page 10
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-52-
Training LM technique in diabetic drugs data was shown in Figure 10:
Figure10. Levenberg-Marquardt of Neural Networks in Diabetic Drugs
The results of implementing LM technique declared that the performance (mean
squared error) of Diabetic drugs is (149582.28294).
Regression(R) is values measure the correlation between outputs and targets. An R value of
1 means a close relationship, 0 a random relationship [29]. Regression of LM technique in
Diabetic drugs was shown in Fig.11:
Figure11. Levenberg-Marquardt of Neural Networks in Diabetic Drugs about Regression
The regression between values is (0.99553) which mean random relationship between
values.
2. Applying Bayesian Regularization Technique in Diabetic Drugs:
This technique typically takes more time, but can result in good generalization for
difficult, small or noisy datasets. Training stops according to adaptive weight minimization
(regularization) [28].
Training Bayesian Regularization technique in diabetic drugs data was shown in Figure12:
Page 11
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-53-
Figure12. Bayesian Regularization of Neural Networks in Diabetic Drugs
The results of implementing Bayesian Regularization technique declared that the
performance (mean squared error) of Diabetic drugs after training is (306.10614).
Regression of Bayesian Regularization technique in diabetic drugs data was shown in
Figure13:
Figure13. Bayesian Regularization of Neural Networks in Diabetes Drugs about Regression
The regression between values is (0.99999) which mean random relationship between values.
3. Scaled Conjugate Gradient Technique in Diabetic drugs:
This technique takes less memory. Training automatically stops when generalization stops
improving, as indicated by an increase in the mean square error of the validation samples [30].
Training Scaled Conjugate Gradient technique in diabetic drugs data was shown in Figure14:
Page 12
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-54-
Figure14. Scaled Conjugate Gradient of Neural Networks in Diabetic Drug
The results of implementing Scaled Conjugate Gradient technique declared that the
performance (mean squared error) of Diabetic drugs after training is (260490.75382).
Regression of Scaled Conjugate Gradient technique in diabetic drugs data was shown in
Fig.15:
Figure15. Scaled Conjugate Gradient of Neural Networks in Diabetic Drugs about Regression
The regression between values is (0.99122) which mean random relationship between values.
3.4.2. Applying Neural Networks in Hypertension Drugs
1. Applying LM Technique in Hypertension Drugs Data
Training LM technique in hypertension drugs data was shown in Figure16:
Page 13
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-55-
Figure 16. Levenberg-Marquardt of Neural Networks in Hypertension Drugs
The results of implementing LM technique declared that the performance (mean squared
error) of Diabetic drugs is (5806.70695).
Regression of LM technique in hypertension drugs data was shown in Figure17:
Figure17. Levenberg-Marquardt of Neural Networks in Hypertension Drugs about Regression
The regression between values is (0.99966) which mean random relationship between
values.
2. Applying Bayesian Regularization Technique in Hypertension Drugs Data:
Training Bayesian Regularization technique in hypertension drugs data was shown in
Figure18:
Page 14
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-56-
Figure18. Bayesian Regularization of Neural Networks in Hypertension Drugs
The results of implementing Bayesian Regularization technique declared that the
performance (mean squared error) of Diabetic drugs is (426.43748).
Regression of Bayesian Regularization technique in hypertension drugs data was shown in Figure19:
Figure19. Bayesian Regularization of Neural Networks in Hypertension Drugs about Regression
The regression between values is (0.99998) which mean random relationship between values.
3. Applying Scaled Conjugate Gradient Technique in Hypertension Drugs Data:
Training Scaled Conjugate Gradient technique in pressure drugs data was shown in Figure20:
Figure20.Scaled Conjugate Gradient of Neural Networks in Hypertension Drugs
Page 15
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-57-
The results of implementing Scaled Conjugate Gradient technique declared that the
performance (mean squared error) of Diabetic drugs is (6515.51252).
Regression of Scaled Conjugate Gradient technique in pressure drugs data was shown in
Figure21:
Figure 21.Scaled Conjugate Gradient of Neural Network in Hypertension Drugs about Regression
The regression between values is (0.99955) which mean random relationship between
values.
4.4 Comparison between Applied Techniques
The researcher made a comparison between all the applied techniques to focus on the
technique with mean square error that shown in Table 4.
Table4. Comparison between Applied Techniques
Applying
neural time series
Neural network
Time series Levenberg
Marquaradt
Bayesian
Regularization
Scaled conjugate
gradient
Diabetic drugs 149582.28294 306.10614 260490.75382 470074.75
Hypertension
drugs 5806.70695 426.43748 6515.51252 184188.3363
The Neural Networks and time series techniques are applying in Diabetic drugs data to
know the MSE for each technique. First, the results of applying Neural Networks techniques
present the MSE of LM technique was (149582.28294 ( ,the MSE of Bayesian Regularization
technique was (306.10614) and the MSE of Scaled conjugate gradient technique was
(260490.75382). Second, the MSE of time series technique was (470074.75). By applying
Neural Networks techniques and time series technique in Hypertension drugs data to knowing
the MSE for each technique, there are many results. First, the results of applying Neural
Networks techniques present the MSE of LM technique was (5806.70695 ( ,the MSE of
Bayesian Regularization technique was (426.43748) and the MSE of Scaled conjugate gradient
technique was (6515.51252). Second, the MSE of time series technique was (184188.3363).
The data showed that time series technique was not best technique in prediction with
Diabetic and Hypertension drugs data because the data was nonlinear and non-smoothly. So
time series is preserved under a smooth change of coordinates is trajectory crossing (or lack
thereof). As shown in results of implementation, a deterministic system once enough lags are
Page 16
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-58-
used the dynamics will lie on a single-valued surface. Forecasting in the case of smooth data
reduces to modeling the shape of the surface so it is not effective with non-smooth data.
In addition to the Bayesian Regularization was the best technique in Diabetic and
Hypertension drugs data because it is searching for hidden correlation relationship that makes
data smooth. BR training aims to minimize the sum of mean squared errors although it takes
time to get results. So Neural Networks are capable of fitting linear and nonlinear functions
without the need for knowing the shape of the underlying function so that it is more suitable for
nonlinear time series prediction.
5. Conclusion ,Recommendation and Future Work
This paper proposed a framework to enhance decisions of distributed system in
Egyptian company for medicine trade. The selection of an effective prediction technique may
be based on comparative tests which cover many forecasting techniques such as Neural
Networks and time series.
The conclusions can be summarized as follows:
1. Tests and comparisons between a number of techniques for both Neural Networks and
time series model resulted in identifying the Neural Networks is the best sales prediction
technique for both drugs based on the least mean square error.
2. Applying the technique of time series for data mining in diabetic and hypertension
drugs data showed that the time series was not efficient with nonlinear and non-
smoothly data .
3. Adopting the techniques of Neural Networks for data mining is more efficient with
nonlinear data that have no correlation with no smooth data. By implementing diabetic
and hypertension drugs data, the results shown the best performance for both drugs data
in Bayesian Regularization technique which gives best prediction of sales.
Accordingly, it is recommended to apply the proposed model so as to enhance decisions
of distribution systems in pharmaceutical companies to decrease the medicine industry cost and
increase the productivity.
The researcher recommends the following as future work:
1. Build forecasting model by using NNs technique in Egypt to decrease the expired
medicines so it will also help in the sustainable development of the pharmaceutical
sector.
2. Developing a proposed web based framework for feedback evaluation.
References
[1]. "World Health Organization," 2018. [Online]. Available:
http://www.emro.who.int/egy/programmes/noncommunicable-diseases.html.
[2]. M. D. B. M. P. Crescenzio Gallo, Data Warehouse Design and Management:Theory and
Practice, Italy, 2010.
[3]. A. N. a. S. Al-Shammari, "The Gmuware Biomedical Data Warehouse System for
Integrating Heterogeneous Biomedical Data Sources," Egyptian Computer Science
Journal, vol. 31, no. 1, Jun 2009.
Page 17
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-59-
[4]. L. Campbell, "An EDW remedies headaches in the pharmaceutical industry," 2015.
[Online]. Available:
http://apps.teradata.com/tdmo/v08n02/Viewpoints/IndustryInsights/PrescriptionForSucce
ss.aspx.
[5]. V. H. H. a. D.-R. L. Hani Omar, "A Hybrid Neural Networks Model for Sales Forecasting
Based on ARIMA and Search Popularity of Article Titles," Computational Intelligence
and Neuroscience journal, 2016.
[6]. T. W. B. Group, " The World Bank," 2017. [Online]. Available:
https://data.worldbank.org/indicator/SH.STA.DIAB.ZS.
[7]. D. M. Hasan, "Hypertension in Egypt: A Systematic Review," vol. 10, 2014.
[8]. "The Ministry of Health," [Online]. Available: http://www.mohp.gov.eg/.
[9]. M. M. S. H. F. Neda Khalil Zadeh, "Intelligent Sales Prediction for Pharmaceutical
Distribution Companies: A Data Mining Based Approach," HINDAWI, 2014.
[10]. S. M. a. M. M. Rouzbeh Ghousi, "Application of Data Mining Techniques in Drug
Consumption Forecasting to Help Pharmaceutical Industry Production Planning," in
International Conference on Industrial Engineering and Operations Management,
Istanbul,Turky, 2012.
[11]. M. S. U. Z. a. M. R. H. Emrana Kabir Hashi, "An Expert Clinical Decision Support
System to Predict Disease Using Classification Techniques," in International Conference
on Electrical, Computer and Communication Engineering (ECCE), Bangladesh, 2017.
[12]. A. H. D. R. a. M. K. Jing Lu, "Data Mining Techniques in Health Informatics: A Case
Study from Breast Cancer Research," in International Conference on Information
Technology in Bio- and Medical Informatics, United Kindom, 2015.
[13]. M. A. A. a. M. K. Jain, "Data Mining Techniques for the Prediction of Kidney Diseases
and Treatment," International Journal Of Engineering And Computer Science, vol. 6, no.
2, pp. 20376-20378, 2017.
[14]. M. S. ,. A. G. a. K. K. Sajida Perveena, "Performance Analysis of Data Mining
Classification Techniques to Predict Diabetes," ELSEVIER, 2016.
[15]. D. O. E. a. A. N. Eldeen, "The Technology Of Using A Data Warehouse To Support
Decision-Making In Health Care," International Journal of Database Management
Systems, 2013.
[16]. R. Dutta, "HEALTH CARE DATA WAREHOUSE SYSTEM ARCHITECTURE FOR
INFLUENZA (FLU)DISEASES," Department of Computer Science & Engineering,
Global Institute of Management & Technology, 2013.
[17]. Jeremiah, "BRAND DEVELOPMENT PROCESS VS. FRAMEWORK," [Online].
Available: http://jeremiahgardner.com/blog/brand-development-process-vs-framework/.
[18]. D. B. a. D. T. Bourgeois, "Chapter 4: Data and Databases," in Information Systems For
Business and Beyond, 2013.
[19]. U. S. B. R. M. A. S. A. K. A. A. a. Q. J. Muhammad Bilal Shahid, "Application of Data
Warehouse in Real Life: State-ofthe-art Survey from User Preferences’ Perspective,"
International Journal of Advanced Computer Science and Applications, 2016.
Page 18
Egyptian Computer Science Journal Vol. 43 No.2 May 2019 ISSN-1110-2586
-60-
[20]. R. M. H. a. M. I. M. Yehia M. K. Helmy, "A Survey on Data Warehouse Quality,"
Egyptian Computer Science Journal, vol. 26, no. 2, Jul 2004.
[21]. S. Greengard, "Organizing data and information," in Information Technology Concept, 2000.
[22]. T. Peng, "A FRAMEWORK FOR DATA CLEANING IN DATA WAREHOUSES," 2003.
[23]. A. T. J. a. L. Tay, "Introduction to Time Series Analysis for Organizational Research:
Methods for Longitudinal Analyses," SAGE, 2017.
[24]. J. Moreno, "Artificial neural networks applied to forecasting time series," Psicothema,
vol. 23, 2011.
[25]. S. S. a. S. R. G. Mahalakshmi, "A survey on forecasting of time series data," in
International Conference on Computing Technologies and Intelligent Data Engineering
(ICCTIDE'16), Kovilpatti, India, 2016.
[26]. B. S. a. P. K. Venugopalan, "Comparison of Neural Networks Training Functions for
Hematoma Classification in Brain CT Images," IOSR Journal of Computer Engineering,
vol. 16, 2013.
[27]. H. P. Gavin, "the Levenberg-Marquardt method for nonlinear least squares curve-fitting
problems," 2017.
[28]. H. Okut, "Bayesian Regularized Neural Networks for Small and Big Data," INTECH, 2016.
[29]. S. E. M. A. a. H. F. Alaa F. Sheta, "A Comparison between Regression, Artificial Neural
Networks and Support Vector Machines for Predicting Stock Market Index," International
Journal of Advanced Research in Artificial Intelligence, vol. 4, 2015.
[30]. H. L. J. W. &. Y. T. Zhongbo Sun, "Two modified spectral conjugate gradient methods
and their global convergence for unconstrained optimization," International Journal of
Computer Mathematics, 2015.